CN102496000B - Urban traffic accident detection method - Google Patents

Abstract

The invention discloses an urban traffic accident detection method. Corresponding moving direction diagrams are built for all frames of video images through optical flow information of a moving object is extracted from a traffic video stream, a traffic monitoring video stream is converted into a moving direction diagram energy sequence through calculating the energy of the moving direction diagram, and whether a traffic accident occurs in the video stream is determined through detecting whether sudden energy increase occurs in the moving direction diagram on a timer shaft. By adopting the method, the characteristic that the local moving direction of at least one moving object suddenly changes in case of collision between moving objects in a traffic accident is utilized to calculate the energy of the moving direction diagram through building the moving direction diagram, and the fluctuation of energy of the moving direction diagram is monitored on the timer shaft to detect traffic accidents, therefore, on one hand, the detection complexity is lowered, and on the other hand, the detection stability and adaptability are improved.

Description

A kind of urban traffic accident detection method

Technical field

The invention belongs to technical field of video monitoring, be specifically related to a kind of detection method of urban traffic accident.

Background technology

Be accompanied by expanding economy, each metropolitan vehicle guaranteeding organic quantity climbs up and up, and urban traffic environment goes from bad to worse, and has more caused Urban Road Traffic Accidents to take place frequently.Traffic congestion, property loss not only can be caused in the urban traffic accident, and more seriously entail dangers to civic life security, cause the loss that can't retrieve.In order to strengthen the managerial ability to urban transportation, the loss of the people's lives and properties that the reduction Urban Road Traffic Accidents causes, major cities are all set up the urban traffic road video surveillance network that covers whole city's traffic route successively, carry out real-time urban transportation monitoring by monitor network, the traffic hazard that occurs is disposed timely.being based upon of urban transportation monitor network reduced the loss that traffic hazard causes to a certain extent, but what adopt due to each city traffic Surveillance center is traditional personal monitoring's means, restricted greatly the development of urban transportation monitor network, at first the personal monitoring can't adapt to large-scale urban transportation monitor network, no matter from cost and benefit, the personal monitoring has its obstacle that can't overcome, and in large-scale monitoring network, this obstacle is particularly evident, secondly traffic hazard often occurred in one second, uncontrollable factor due to the personal monitoring, make it under the environment of extensive traffic monitoring, accuracy rate and stability can't be protected.Shortcoming for the personal monitoring, people have adopted the intelligent video analysis technology that the video flowing of traffic monitoring is analyzed, and then detect and whether to have traffic hazard to occur, and existing traffic hazard detection technique ubiquity based on video flowing that adaptive capacity to environment is poor, a little less than anti-noise ability and detection speed wait slowly deficiency.

" a kind of urban traffic accident automatic identifying method and system " disclosed at publication number CN 102073851A, the method is when carrying out vehicle tracking, use vehicle center and color as the feature of vehicle target, utilize the camshift algorithm to upgrade current tracking queue, use next vehicle center constantly of kalman filter forecasting, and send the vehicle center of prediction to the camshift algorithm, when traffic hazard is identified, with the velocity variations of extracting, horizontal level changes, the upright position changes and direction of motion changes the weighting system of multiply by separately, then summation, if the numerical value that obtains is greater than the accident threshold value, for traffic hazard occurs, otherwise be normal condition.The method mainly uses the track of moving vehicle to carry out the detection of traffic hazard, but the track of moving vehicle easily occurs incomplete and crosses, do not have stability in complicated traffic environment, at night, vehicle is followed the tracks of difficulty larger, and the accident threshold value is difficult for determining, directly have influence on the detection effect, use simultaneously the method for multiple goal vehicle tracking more consuming time, be difficult to carry out the real-time processing of SD video.

" a kind of automatic testing method of vehicle traffic accident " disclosed at publication number CN 101105892A, the method is analyzed continuously to video image, measure the relative movement speed of mobile object in video image, in video image, more than one object stops in movement fast if detect, and the static duration surpass certain-length, static object is inferred to be the accident of having occured.Stipulate the range of size of specific mobile object, this is of a size of the relative size in video pictures, the range of size different according to mobile object, the type of object but automatic decision has an accident.The detection of the method prospect of the application and object tracking technology are carried out the detection of traffic hazard, the rule of its setting is comparatively simple, meeting for escaping behavior after traffic accident occurs undetected, and can carry out false retrieval for the parking maintenance of vehicle, do not have stability under complicated traffic environment, and detection threshold is difficult definite, simultaneously more consuming time based on the method for vehicle tracking, is difficult to carry out the real-time processing of SD video.

Summary of the invention

The objective of the invention is to have proposed a kind of detection method of urban traffic accident in order to solve the above-mentioned shortcoming of existing urban traffic accident detection method existence.

Technical scheme of the present invention is: a kind of detection method of urban traffic accident comprises the steps:

S1. the Traffic Surveillance Video image is carried out pre-service and extract the Optic flow information of video image;

S2. build direction of motion figure corresponding to each two field picture according to the Optic flow information that extracts;

S3. rely on the direction of motion figure that builds, calculate the energy of direction of motion figure, make continuous Traffic Surveillance Video circulation become continuous direction of motion figure energy sequence;

S4. according to the direction of motion figure energy sequence that gets, carry out the urban traffic accident and detect.

Further, step S1 specifically comprises step by step following:

S11. the Traffic Surveillance Video image is carried out medium filtering, the noise that exists in the filtering video image;

S12. optical flow computation is carried out in the Traffic Surveillance Video image of filtering, extract the Optic flow information of moving object in video image;

Each light stream that S13. will obtain uses the mode of (starting point coordinate, terminating point coordinate) to store, and forms the Optic flow information set of this frame video image.

Further, step S2 specifically comprises step by step following:

S21. be that each Optic flow information builds corresponding rectangle agglomerate, take out successively every a pair of coordinate points in the Optic flow information set that step S13 obtains, with the lower left corner coordinate of starting point coordinate as the rectangle agglomerate, with the upper right corner coordinate of terminating point coordinate as the rectangle agglomerate, build the rectangle agglomerate;

S22. calculate the direction of each light stream, take out successively every a pair of coordinate points in the Optic flow information set that obtains in step S13, utilize coordinate Calculation starting point and the line of terminating point and the angle of transverse axis of starting point and terminating point, this angle is the direction of this light stream.

S23. according to the light stream direction of calculating, for each rectangle agglomerate carries out assign operation;

Whether whether S24. detect the rectangle agglomerate and occur intersecting, occurred intersecting according to the coordinate Calculation of rectangle agglomerate and other rectangle agglomerate, the rectangle agglomerate that intersects has appearred in record;

S25. will intersect the rectangle agglomerate as the new connected domain of a unification, agglomerate exists and these rectangle agglomerates that intersect are not re-used as independently; The rectangle agglomerate that does not have to occur to intersect is regarded as independently connected domain, and the rectangle agglomerate that occurs to intersect forms new connected domain;

S26. the rectangle agglomerate that intersects is carried out mixing operation, if two rectangle agglomerates occur to intersect, the pixel value in two rectangle agglomerate disjoint zones remains unchanged, and the pixel value of intersecting area is the average of two rectangle agglomerate brightness values;

S27. with all connected domains, according to its position, coverage and pixel value, be shown in the new images with Traffic Surveillance Video image equal proportion, this image is direction of motion figure;

S28. build the direction of motion graphic sequence.Build corresponding direction of motion figure for each frame video image according to step S21～S27, thereby change sequence of video images into the direction of motion graphic sequence.

Further, step S3 specifically comprises step by step following:

S31. calculate the internal energy of each connected domain of direction of motion figure, at first calculate the information entropy in this connected domain, then information entropy and this connected domain brightness value species number are multiplied each other, obtain the within value of this connected domain.

S32. calculate the external energy of each connected domain of direction of motion figure, the external energy of specific connected domain is the summation of absolute value of the difference of the internal energy of this connected domain and all the other connected domain internal energies;

S33. calculate the energy of each connected domain in direction of motion figure, the energy of each connected domain is the summation of internal energy and the external energy of this connected domain;

S34. search the highest connected domain of energy value in this direction of motion figure, and record this maximum energy value;

S35. calculate the average of all the other all the connected domain energy values except the connected domain of energy value maximum;

S36. determine the energy of direction of motion figure.If the maximum energy value integral multiple is in the average energy value of all the other connected domains, the energy value of this direction of motion figure is maximum energy value, if the integral multiple of not enough all the other the connected domain the average energy value of maximum energy value, the energy value of this direction of motion figure is the average energy value of all connected domains;

S37. build direction of motion figure energy sequence.To the operation of each the direction of motion figure in the direction of motion graphic sequence according to step S31～S36, from obtaining the energy of each direction of motion figure, form direction of motion figure energy sequence.

Further, step S4 specifically comprises step by step following:

S41. according to normal direction of motion figure energy sequence, calculate under this traffic scene the average of energy hunting Gaussian distribution;

S42. according to normal direction of motion figure energy sequence, calculate under this traffic scene the standard deviation of energy hunting Gaussian distribution;

S43. according to the fiducial interval of determining, utilize the characteristic of Gaussian distribution and average and the standard deviation that step S41～S42 calculates, obtain the upper bound of direction of motion figure energy hunting value.

S44. according to the upper bound of direction of motion figure energy hunting value, detect whether the urban traffic accident occurs, if the energy value of a certain direction of motion figure over the upper bound, is estimated to be, traffic hazard has occured;

S45. determine the suspicious region that traffic hazard occurs, if infer, traffic hazard has occured, seek the position of the connected domain of energy value maximum in the corresponding direction of motion figure of this video frame image, the suspicious region that occurs for traffic hazard is demarcated in this position.

Beneficial effect of the present invention: method of the present invention is by the Optic flow information of the moving object of extraction from traffic video stream, for every frame video image builds corresponding direction of motion figure, by the energy that calculates direction of motion figure, the Traffic Surveillance Video circulation is become direction of motion figure energy sequence, thereby by detect energy whether occurring and suddenly increase to determine in video flowing, whether traffic hazard to have occured in direction of motion figure energy sequence on time shaft.The collision of method utilization of the present invention moving object in traffic hazard can cause the direction of motion of the appearance locality of at least one moving object this characteristics of suddenling change, by building direction of motion figure, calculate the energy of direction of motion figure, energy hunting monitors to detect traffic hazard to direction of motion figure on time shaft, reduce on the one hand the complicacy that detects, improved on the other hand the Stability and adaptability that detects.

Description of drawings

Fig. 1 is urban traffic accident detection method schematic flow sheet of the present invention.

Fig. 2 is that schematic diagram is detected in urban traffic accident of the present invention.

Fig. 3 is the direction of motion figure schematic diagram of various vehicle behaviors in the embodiment of the present invention.

Fig. 4 is direction of motion figure energy sequence schematic diagram in the embodiment of the present invention.

Fig. 5 is testing result schematic diagram in the embodiment of the present invention.

Embodiment

The present invention is described in further detail below in conjunction with accompanying drawing and specific embodiment:

The detection method of urban traffic accident of the present invention specifically comprises the steps: as shown in Figure 1

S1. the Traffic Surveillance Video image is carried out pre-service and extract the Optic flow information of video image;

S2. build direction of motion figure corresponding to each two field picture according to the Optic flow information that extracts;

S3. rely on the direction of motion figure that builds, calculate the energy of direction of motion figure, make continuous Traffic Surveillance Video circulation become continuous direction of motion figure energy sequence;

S4. according to the direction of motion figure energy sequence that gets, carry out the urban traffic accident and detect.

In the process that this programme carries out detecting the urban traffic accident, at first the Traffic Surveillance Video image being carried out anti-noise processes, reduce noise to the impact of video analysis, then extract the Optic flow information in video image, use light stream to describe the movable information of object in video image.

Second step relies on the Optic flow information that extracts from traffic video image, for each two field picture builds corresponding direction of motion figure.At first, according to position, direction and the mould value of each light stream, build the relevant parameter (position, direction and mould value) that the rectangle agglomerate represents light stream; Secondly, according to position and the size of rectangle agglomerate, detect each rectangle agglomerate and whether occur intersecting; At last, the rectangle agglomerate that intersects is merged, complete the structure of direction of motion figure.

The 3rd step relied on to build the direction of motion figure that completes, and calculated the energy of direction of motion figure.At first, calculate the energy value of each connected domain in direction of motion figure, the energy of any one connected domain is the summation of internal energy and external energy.Secondly, energy according to all connected domains that obtain, whether the maximum energy value of judgement doubles the average energy of all the other connected domains, at last, according to judged result, determine the energy value of every frame direction of motion figure, if greater than, the energy value of this direction of motion figure is the maximal value in all connected domain energy, if less than, the energy value of this direction of motion figure is the mean value of all connected domain energy values.Optic flow information according to each frame of video, capital process described according to second step, build corresponding direction of motion figure, then calculate the energy value of direction of motion figure according to this step, therefore continuous Traffic Surveillance Video stream just is converted into continuous direction of motion figure energy sequence.

In the 4th step, according to the direction of motion figure energy sequence that gets, carry out the urban traffic accident and detect.At first, use Gauss model study specific road section direction of motion figure energy amplitude of fluctuation under normal circumstances.Secondly, use the Gauss model that study is completed to detect suddenly increasing of direction of motion figure energy, suddenly increase if small probability occurs, be identified as traffic hazard.At last, according to recognition result, send the suspicious region of reporting to the police and showing accident.

At first the Traffic Surveillance Video image is carried out the Optic flow information of pre-service and extraction video image, its concrete implementation step is as follows:

(1.1) the Traffic Surveillance Video image is carried out medium filtering, the noise that exists in the filtering video image.

(1.2) the Traffic Surveillance Video image of filtering carried out optical flow computation, extract the Optic flow information of moving object in video image.

Each light stream that (1.3) will obtain uses the mode of (starting point coordinate, terminating point coordinate) to store, and forms the Optic flow information set O={o of this frame video image ₁, o ₂..., o _n.Any light stream o in Optic flow information set O _iCan be expressed as

Expression light stream o _iStarting point, its coordinates table is shown

Expression light stream o _iCoordinate points, its coordinates table is shown

Traffic hazard is made of very complicated object of which movement pattern often, under different traffic environments, the mode that traffic hazard occurs often is not quite similar, be difficult to go to portray with unified model, no matter but how complicated and changeable traffic hazard is, it is fundamentally the mutual collision of two moving objects.In traffic hazard, this mutual collision tends to cause the direction of motion of at least one moving object to undergo mutation, and this sudden change does not have globality, turn around from the vehicle under the normal traffic environment, turn inside diameter and vehicle bring to a halt all different, because these motions all have the globality of vehicle movement, therefore to build direction of motion figure be exactly in order can to highlight traffic hazard and to occur the time to this programme, the direction of motion sudden change of locality.The concrete steps that build the corresponding sports directional diagram by the Optic flow information set that utilizes video image are as follows:

(2.1) be that each Optic flow information builds corresponding rectangle agglomerate.Take out successively every a pair of coordinate points in the Optic flow information set that obtains in step 1.3, with the lower left corner coordinate of starting point coordinate as the rectangle agglomerate, with the upper right corner coordinate of terminating point coordinate as the rectangle agglomerate, build the rectangle agglomerate.By light stream o _iThe rectangle agglomerate W that builds _iCan be expressed as:

$W_i@\left\{(x,y)\right|x_i^s<x<x_i^e,y_i^s<y<y_i^e\}.$

(2.2) calculate the direction of each light stream, take out successively every a pair of coordinate points in the Optic flow information set that obtains in step 1.3, utilize coordinate Calculation starting point and the line of terminating point and the angle of x axle of starting point and terminating point, this angle is the direction of this light stream, light stream o _iDirection indication be θ _i

(2.3) according to the light stream direction of calculating, for each rectangle agglomerate carries out assign operation.

The rectangle agglomerate can use the coordinate in the lower left corner and the upper right corner to obtain the light stream direction of this rectangle agglomerate representative according to step 2.2 arbitrarily, the angular range of this light stream direction from 0 to 2 π is mapped to 0 to 255 brightness range, pixel assignment all in this rectangle agglomerate is the brightness value after shining upon.

Any rectangle agglomerate W _iIn pixel value

Determined by following formula:

$I_{W_i}=\left\{\begin{array}{cc}\frac{{\mathrm{\&theta;}}_{i}R}{2\mathrm{\&pi;}}+\frac{\mathrm{<figure-callout\; id="2"\; label="R"\; filenames="HDA0000108025280000011.png"\; state="\{\{state\}\}">R</figure-callout>}}{2}& \mathrm{for}{\mathrm{\&theta;}}_i\&Element;[0,\mathrm{\&pi;}]\\ \frac{({\mathrm{\&theta;}}_i-\mathrm{\&pi;})\mathrm{<figure-callout\; id="2"\; label="R"\; filenames="HDA0000108025280000011.png"\; state="\{\{state\}\}">R</figure-callout>}}{2\mathrm{\&pi;}}& \mathrm{for}{\mathrm{\&theta;}}_i\&Element;(\mathrm{\&pi;},2\mathrm{\&pi;}]\end{array}\right.,$

Wherein, R is the higher limit that needs the brightness range of mapping, R=255 here.

Whether (2.4) detect the rectangle agglomerate occurs intersecting.Whether occurred intersecting according to coordinate Calculation and other rectangle of rectangle agglomerate, the rectangle agglomerate that intersects has appearred in record.

(2.5) will intersect the rectangle agglomerate as the new connected domain of a unification, agglomerate exists and these rectangle agglomerates that intersect are not re-used as independently.The rectangle agglomerate that does not have to occur to intersect is regarded as independently connected domain, and the rectangle agglomerate that occurs to intersect forms new connected domain.New connected domain B _iCan be determined by following formula:

$B_i=\left\{\begin{array}{cc}{W}_i& \mathrm{for}(W_i\&cap;W_j=\mathrm{\&phi;},j=1,...,n)\\ W_i\&cup;W_j& \mathrm{for}(W_i\&cap;W_j\&NotEqual;\mathrm{\&phi;},j=1,...,n)\end{array}\right.$

(2.6) the rectangle agglomerate that intersects is carried out mixing operation.If two rectangle agglomerates occur to intersect, the pixel value in two rectangle agglomerate disjoint zones remains unchanged, and the pixel value of intersecting area is the average of two rectangle agglomerate brightness values.Crossing situation occurs in a plurality of rectangle agglomerates, processes according to the mode of two agglomerates equally.Any connected domain B _iIn pixel value

Can be determined by following mode:

$I_{B_i}=\left\{\begin{array}{cc}{I}_{W_i}& ((x,y)\&Element;W_i,(x,y)\&NotElement;W_i\&cap;W_j)\\ I_{W_j}& ((x,y)\&Element;W_j,(x,y)\&NotElement;W_i\&cap;W_j)\\ I_{W_i\&cap;W_j}& ((x,y)\&Element;W_i\&cap;W_j)\end{array}\right.;$ Wherein, $I_{W_i\&cap;W_j}=\frac{I_{W_i}+{\mathrm{<figure-callout\; id="2"\; label="I\; W\; j"\; filenames="HDA0000108025280000011.png"\; state="\{\{state\}\}">I</figure-callout>}}_{W_j}}{2}.$

(2.7) with all connected domains, according to its position, coverage and pixel value, be shown in the new images with Traffic Surveillance Video image equal proportion, this image is direction of motion figure.

(2.8) build the direction of motion graphic sequence.Build corresponding direction of motion figure for each frame video image execution in step 2.1～2.7, thereby change sequence of video images into direction of motion graphic sequence F={f ¹, f ²..., f ^k... }.

Build by step 2.1～2.7 the direction of motion figure of coming, if there is the motion sudden change of locality, a certain connected domain in this direction of motion figure must exist obvious luminance difference, because different brightness is representing different direction of motion.By the structure of direction of motion figure, the mode of original sparse Optic flow information with connected domain showed, the parameter informations such as the direction that has not only kept light stream and mould value, and also the motion that can highlight locality suddenlys change.In order to weigh the degree of this locality motion sudden change, to calculate one by one each connected domain in direction of motion figure, the zone that bumps, obvious luminance difference appears because the motion sudden change of locality can cause a certain connected domain, be that the information entropy is too high, and the zone that bumps on the other hand necessarily obviously is different from other the connected domain that does not bump, so the energy of each connected domain is comprised of internal energy and external energy.Rely on the energy that calculates each connected domain, can obtain the definite energy of direction of motion figure, the concrete steps of whole operation are as follows:

(3.1) calculate the internal energy of each connected domain of direction of motion figure.At first, calculate the information entropy in this connected domain, then information entropy and this connected domain brightness value species number are multiplied each other, obtain the within value of this connected domain.Any connected domain B _iInternal energy D (B _i) can be determined by following formula:

Wherein, N is connected domain B _iThe species number of middle brightness value, p (x) expression pixel intensity is the probability of x, equaling brightness is that the pixel count of x is divided by the total pixel number of image.

(3.2) calculate the external energy of each connected domain of direction of motion figure.The external energy of specific connected domain is the summation of absolute value of the difference of the internal energy of this connected domain and all the other connected domain internal energies.

(3.3) calculate the energy of each connected domain in direction of motion figure.The energy of each connected domain is the summation of internal energy and the external energy of this connected domain.Any connected domain B _iGross energy E (B _i) can be determined by following formula:

$E\left(B_i\right)=D\left(B_i\right)+\underset{j=1,j\&NotEqual;i}{\overset{m}{\mathrm{\&Sigma;}}}V(B_i,B_j)$

Wherein, V (B _i, B _j)=| D (B _i)-D (B _j) |, m is the sum of connected domain in direction of motion figure.

(3.4) search the connected domain of energy value maximum in this direction of motion figure, and record this maximum energy value, at direction of motion figure f ^kMiddle Energy maximum value is expressed as

(3.5) calculate the average of all the other all the connected domain energy values except the connected domain of energy value maximum.

(3.6) determine the energy of direction of motion figure.If the maximum energy value integral multiple is in the average energy value of all the other connected domains, the energy value of this direction of motion figure is maximum energy value, if the integral multiple of not enough all the other the connected domain the average energy value of maximum energy value, the energy value of this direction of motion figure is the average energy value of all connected domains.Integral multiple in the present embodiment is specially 2 times.

Arbitrary motion directional diagram f ^kENERGY E (f ^k) calculated by following formula:

$E\left(f^k\right)=\left\{\begin{array}{cc}E\left(B_j^k\right)& \mathrm{if\; E}\left(B_j^k\right)>\frac{2\underset{i=1i\&NotEqual;j}{\overset{m}{\mathrm{\&Sigma;}}}E\left(B_i^k\right)}{m-1}\\ \underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}E\left(B_i^k\right)/m& \mathrm{otherwise}\end{array}\right.$

(3.7) build direction of motion figure energy sequence.To the operation of each the direction of motion figure execution in step 3.1～3.6 in the direction of motion graphic sequence, from obtaining the energy of each direction of motion figure, form direction of motion figure energy sequence E (F)={ E (f ¹), E (f ²) ..., E (f ^k) ....

Here, only be to provide a kind of mode of calculating direction of motion figure, those of ordinary skill in the art should recognize, also can calculate direction of motion figure by alternate manner.

In the situation that there is no accident, direction of motion figure energy sequence can fluctuate within the specific limits, the scope of fluctuation is different because of traffic environment, in case accident, suddenly increasing can appear in the direction of motion energy sequence, substantially exceed original normal fluctuation range, by learning the fluctuation range of direction of motion figure energy in normal situation, it can adaptively be the upper bound that different traffic environments are set energy hunting, thereby complete the detection of urban traffic accident, its concrete steps are as follows:

(4.1) according to normal direction of motion figure energy sequence, calculate under this traffic scene the average of energy hunting Gaussian distribution.To arbitrary motion directional diagram f ^kWhen detecting, the average μ of energy hunting Gaussian distribution ^kCalculated by following formula:

${\mathrm{\&mu;}}^k=\frac{1}{T^k}\underset{k=1}{\overset{T^k}{\mathrm{\&Sigma;}}}E\left(f^k\right)$

Wherein, T ^kFor to arbitrary motion directional diagram f ^kWhen detecting, the length of direction of motion figure energy sequence.

(4.2) according to normal direction of motion figure energy sequence, calculate under this traffic scene the standard deviation of energy hunting Gaussian distribution.To arbitrary motion directional diagram f ^kWhen detecting, the standard deviation sigma of energy hunting Gaussian distribution ^kCalculated by following formula:

${\left({\mathrm{\&sigma;}}^k\right)}^2=\frac{1}{T^k}\underset{k=1}{\overset{T^k}{\mathrm{\&Sigma;}}}{(E\left(f^k\right)-{\mathrm{\&mu;}}^k)}^2$

Wherein, T ^kFor to arbitrary motion directional diagram f ^kWhen detecting, the length of direction of motion figure energy sequence.

Need to prove, normal direction of motion figure energy sequence in step (4.1) and (4.2) is to obtain the training stage before detecting, and concrete is is that the monitor video image without traffic hazard that judges obtains according to step S1-S3 with the people.

(4.3) according to the fiducial interval of determining, utilize the characteristic of Gaussian distribution and average and the standard deviation that step 4.1～4.2 calculate, obtain the upper bound of direction of motion figure energy hunting value.To arbitrary motion directional diagram f ^kWhen detecting, the upper bound of direction of motion figure energy hunting value

Specific formula for calculation is:

(4.4) according to the upper bound of direction of motion figure energy hunting value, detect whether the urban traffic accident occurs.If the energy value of a certain direction of motion figure over the upper bound, is estimated to be, traffic hazard has occured.

$\left\{\begin{array}{cc}{f}^k\&Element;S& \mathrm{if\; E}\left(f^k\right)>{\mathrm{\&beta;}}_h^k,\\ f^k\&Element;R& \mathrm{otherwise},\end{array}\right.$

Wherein, S is the traffic hazard set, and R is the normal traffic set, if traffic hazard has occured S ≠ φ.

(4.5) determine the suspicious region that traffic hazard occurs.If traffic hazard has occured, seek the position of the connected domain of energy value maximum in the corresponding direction of motion figure of this video frame image, the suspicious region that occurs for traffic hazard is demarcated in this position.

Schematic diagram is detected as shown in Figure 2 in the urban traffic accident, can be clearly seen that from figure the Optic flow information that extracts is converted into corresponding direction of motion figure from Traffic Surveillance Video, by calculating the energy of direction of motion figure, at time shaft, the energy of direction of motion figure is observed, suddenly increasing of energy occuring be detected as traffic hazard, determines by the connected domain of seeking energy maximum in direction of motion figure the suspicious region that traffic hazard occurs.

The direction of motion figure schematic diagram of various vehicle behaviors as shown in Figure 3, build the direction of motion figure that forms by step 2.1～step 2.7, can demonstrate different picture characteristics to different traffic behaviors, can be clearly seen that in the drawings normally travelling of vehicle, the blocking of vehicle, the turning of vehicle and the formed direction of motion figure of collision of vehicle are distinct, and this also lays the foundation for the detection of traffic hazard.

Direction of motion figure energy sequence schematic diagram as shown in Figure 4, get on to observe the energy hunting of direction of motion figure from time shaft, the corresponding direction of motion figure of frame of video that can see accident has the highest energy value, and connected domain corresponding to its highest energy value zone that occurs of traffic hazard just, and in the scope of normal energy fluctuation, what in direction of motion figure, connected domain corresponding to highest energy value often showed is to monitor moving object active in scene.

Urban traffic accident testing result exploded view as shown in Figure 5, as can be seen from the figure, no matter daytime or night, or under the complicated traffic environments such as highway or crossroad, technical scheme of the present invention has all obtained gratifying result.

The collision of method utilization of the present invention moving object in traffic hazard can cause the direction of motion of the appearance locality of at least one moving object this characteristics of suddenling change, by building direction of motion figure, calculate the energy of direction of motion figure, energy hunting monitors to detect traffic hazard to direction of motion figure on time shaft, reduce on the one hand the complicacy of detection algorithm, improved on the other hand the Stability and adaptability that detects, compare with existing method, method of the present invention has higher accuracy and dirigibility.

Those of ordinary skill in the art will appreciate that, embodiment described here is in order to help reader understanding's principle of the present invention, should to be understood to that protection scope of the present invention is not limited to such special statement and embodiment.Those of ordinary skill in the art can make various other various concrete distortion and combinations that do not break away from essence of the present invention according to these technology enlightenments disclosed by the invention, and these distortion and combination are still in protection scope of the present invention.

Claims (4)

1. the detection method of a urban traffic accident, is characterized in that, comprises the steps:

S1. the Traffic Surveillance Video image is carried out pre-service and extract the Optic flow information of video image;

S2. build direction of motion figure corresponding to each two field picture according to the Optic flow information that extracts;

S3. rely on the direction of motion figure that builds, calculate the energy of direction of motion figure, make continuous Traffic Surveillance Video circulation become continuous direction of motion figure energy sequence;

Step S3 specifically comprises step by step following:

S31. calculate the internal energy of each connected domain of direction of motion figure, at first calculate the information entropy in this connected domain, then information entropy and this connected domain brightness value species number are multiplied each other, obtain the within value of this connected domain;

S32. calculate the external energy of each connected domain of direction of motion figure, the external energy of specific connected domain is the summation of absolute value of the difference of the internal energy of this connected domain and all the other connected domain internal energies;

S33. calculate the energy of each connected domain in direction of motion figure, the energy of each connected domain is the summation of internal energy and the external energy of this connected domain;

S34. search the highest connected domain of energy value in this direction of motion figure, and record this maximum energy value;

S35. calculate the average of all the other all the connected domain energy values except the connected domain of energy value maximum;

S36. determine the energy of direction of motion figure, if the maximum energy value integral multiple is in the average energy value of all the other connected domains, the energy value of this direction of motion figure is maximum energy value, if the integral multiple of not enough all the other the connected domain the average energy value of maximum energy value, the energy value of this direction of motion figure is the average energy value of all connected domains;

S37. build direction of motion figure energy sequence, to the operation of each the direction of motion figure in the direction of motion graphic sequence according to step S31 ~ S36, from obtaining the energy of each direction of motion figure, form direction of motion figure energy sequence;

S4. according to the direction of motion figure energy sequence that gets, carry out the urban traffic accident and detect.

2. the detection method of urban traffic accident according to claim 1, is characterized in that, step S1 specifically comprises step by step following:

S11. the Traffic Surveillance Video image is carried out medium filtering, the noise that exists in the filtering video image;

S12. optical flow computation is carried out in the Traffic Surveillance Video image of filtering, extract the Optic flow information of moving object in video image;

Each light stream that S13. will obtain uses the mode of (starting point coordinate, terminating point coordinate) to store, and forms the Optic flow information set of this frame video image.

3. the detection method of urban traffic accident according to claim 2, is characterized in that, step S2 specifically comprises step by step following:

S21. be that each Optic flow information builds corresponding rectangle agglomerate, take out successively every a pair of coordinate points in the Optic flow information set that step S13 obtains, with the lower left corner coordinate of starting point coordinate as the rectangle agglomerate, with the upper right corner coordinate of terminating point coordinate as the rectangle agglomerate, build the rectangle agglomerate;

S22. calculate the direction of each light stream, take out successively every a pair of coordinate points in the Optic flow information set that obtains in step S13, utilize coordinate Calculation starting point and the line of terminating point and the angle of transverse axis of starting point and terminating point, this angle is the direction of this light stream;

S23. according to the light stream direction of calculating, for each rectangle agglomerate carries out assign operation;

Whether whether S24. detect the rectangle agglomerate and occur intersecting, occurred intersecting according to the coordinate Calculation of rectangle agglomerate and other rectangle agglomerate, the rectangle agglomerate that intersects has appearred in record;

S25. will intersect the rectangle agglomerate as the new connected domain of a unification, agglomerate exists and these rectangle agglomerates that intersect are not re-used as independently; The rectangle agglomerate that does not have to occur to intersect is regarded as independently connected domain, and the rectangle agglomerate that occurs to intersect forms new connected domain;

S26. the rectangle agglomerate that intersects is carried out mixing operation, if two rectangle agglomerates occur to intersect, the pixel value in two rectangle agglomerate disjoint zones remains unchanged, and the pixel value of intersecting area is the average of two rectangle agglomerate brightness values;

S27. with all connected domains, according to its position, coverage and pixel value, be shown in the new images with Traffic Surveillance Video image equal proportion, this image is direction of motion figure;

S28. build the direction of motion graphic sequence, build corresponding direction of motion figure for each frame video image according to step S21 ~ S27, thereby change sequence of video images into the direction of motion graphic sequence.

4. the detection method of according to claim 2 or 3 described urban traffic accident, is characterized in that, step S4 specifically comprises step by step following:

S41. according to normal direction of motion figure energy sequence, calculate under this traffic scene the average of energy hunting Gaussian distribution;

S42. according to normal direction of motion figure energy sequence, calculate under this traffic scene the standard deviation of energy hunting Gaussian distribution;

S43. according to the fiducial interval of determining, utilize the characteristic of Gaussian distribution and average and the standard deviation that step S41 ~ S42 calculates, obtain the upper bound of direction of motion figure energy hunting value;

S44. according to the upper bound of direction of motion figure energy hunting value, detect whether the urban traffic accident occurs, if the energy value of a certain direction of motion figure over the upper bound, is estimated to be, traffic hazard has occured;

S45. determine the suspicious region that traffic hazard occurs, if infer, traffic hazard has occured, seek the position of the connected domain of energy value maximum in the corresponding direction of motion figure of this video frame image, the suspicious region that occurs for traffic hazard is demarcated in this position.

Images